WWW 2023: Event prediction using case-based reasoning over knowledge graphs

Key idea: typical LP models are ill-suited for this task as they are incapable of performing inductive link prediction for new, unseen event entities and they require retraining as knowledge is added or changed in the underlying KG. This paper uses statistical measures to identify similar events and performs path-based predictions, requiring no training step.

1. Abstract
   1. Task: predict properties about new consequent events based on similar cause-efect events present in the KG
   2. Methodology: Frame the task to a 2-hop LP task, where the first hop is a causal relation connecting a cause event to a new effect event and the second hop is a property about the new event which we wish to predict
   3. Dataset: The model is applied on a novel dataset containing newsworthy events with causal relations curated from Wikidata, where EvCBR outperforms baselines in- cluding translational-distance-based, GNN-based, and rule-based LP models.
2. Intro
   1. Motivation: A major limitation of most LP methods is that relations can only be predicted among entities which already exist in KG
   2. Key idea: leverage examples of past events in the KG to perform LP for properties of unseen entities.
   3. Methodology: We frame our problem as a 2-hop link prediction task – e.g., starting from a cause event, we assume the existence of a causal relation to a new efect event and predict its properties
   4. Contribution summary
      1. leveraging knowledge about similar cause effect events to make predictions about the unseen effect
      2. curate and release a novel dataset surrounding causal events in Wikidata, extracting news events that are con- nected by causal relations as well as their local connections.
      3. our model shows superior performance on our event dataset as well as com- petitive performance on a modifed evaluation dataset based on the FB15k-237 dataset.
3. Experiments
   1. Problem formulation
      1. Predict missing tail which is not in the knowledge graph
   2. Case retrieval
      1. Entity similarity: For each entity, formulates a vector contains the number of entities dimensions